On Thu, 14 May 2026 16:26:25 +0100
Steven Price <[email protected]> wrote:

> On 12/05/2026 12:37, Boris Brezillon wrote:
> > This is not a bug per-se, because this lock is never taken in an
> > interrupt context, but it's not consistent with the other users of this
> > lock. We're also planning on transitioning GPU event processing to
> > a hard handler. Again, this alone wouldn't justify using the IRQ-safe
> > variant, because then this _lock/unlock sequence would be in the
> > hard-IRQ path, where IRQs are already disabled, but let's do it anyway,
> > to keep things consistent.
> > 
> > While at it, transition to a guard() instead of a plain lock/unlock
> > sequence.
> > 
> > Signed-off-by: Boris Brezillon <[email protected]>  
> 
> Reviewed-by: Steven Price <[email protected]>
> 
> Although this now leaves the lock in panthor_pwr_irq_handler() the odd
> one out...

Yeah, I realized I had overlooked the PWR IRQ after sending v2. I'll
transition that one to an hard handler in v3, and use the irqsave
variant there as well.

> 
> Thanks,
> Steve
> 
> > ---
> >  drivers/gpu/drm/panthor/panthor_gpu.c | 3 +--
> >  1 file changed, 1 insertion(+), 2 deletions(-)
> > 
> > diff --git a/drivers/gpu/drm/panthor/panthor_gpu.c 
> > b/drivers/gpu/drm/panthor/panthor_gpu.c
> > index d0be758ea3e1..b9c51f8a051d 100644
> > --- a/drivers/gpu/drm/panthor/panthor_gpu.c
> > +++ b/drivers/gpu/drm/panthor/panthor_gpu.c
> > @@ -110,12 +110,11 @@ static void panthor_gpu_irq_handler(struct 
> > panthor_irq *pirq, u32 status)
> >     if (status & GPU_IRQ_PROTM_FAULT)
> >             drm_warn(&ptdev->base, "GPU Fault in protected mode\n");
> >  
> > -   spin_lock(&ptdev->gpu->reqs_lock);
> > +   guard(spinlock_irqsave)(&ptdev->gpu->reqs_lock);
> >     if (status & ptdev->gpu->pending_reqs) {
> >             ptdev->gpu->pending_reqs &= ~status;
> >             wake_up_all(&ptdev->gpu->reqs_acked);
> >     }
> > -   spin_unlock(&ptdev->gpu->reqs_lock);
> >  }
> >  
> >  static irqreturn_t panthor_gpu_irq_threaded_handler(int irq, void *data)
> >   
> 

Reply via email to